Evidence of differences in diurnal electrodermal, temperature and heart rate patterns by mental health status in free-living data

Daniel McDuff, Isaac Galatzer-Levy, Seamus Thomson, Andrew Barakat, Conor Heneghan, Samy Abdel-Ghaffar, Jacob Sunshine, Ming-Zher Poh, Lindsey Sunden, John B Hernandez, Allen Jiang, Xin Liu, Ari Winbush, Benjamin Nelson, Nicholas B Allen
BMJ Mental Health
Google, Seattle, Washington, USA

First Page Preview

First page preview

Table of Contents

Overall Summary

Study Background and Main Findings

This study investigates the relationship between passively collected physiological data from a commercial smartwatch (Fitbit Sense 2) and self-reported mental health symptoms in a community sample of 237 participants. The research aims to address the limitations of traditional, lab-based studies on electrodermal activity (EDA), a measure of sympathetic nervous system arousal, by leveraging the continuous monitoring capabilities of wearable technology in a free-living context. Over a four-week period, the study collected EDA, heart rate, skin temperature, and step count data, alongside self-reported measures of depression, anxiety, and perceived stress.

The primary finding is that individuals with elevated depression and anxiety symptoms exhibited significantly higher tonic EDA, skin temperature, and heart rate compared to those without these symptoms. This difference in EDA was most pronounced during the early morning hours. Importantly, the elevated EDA in the depressed group was not associated with increased physical activity, suggesting that the heightened sympathetic arousal is not simply a byproduct of behavior. The study's control group showed similar diurnal patterns to a larger, independent dataset of Fitbit users, validating the baseline measurements. While the depressed and anxious groups showed similar physiological patterns, elevated stress was only associated with higher skin temperature.

The researchers used a non-linear cosinor model to analyze the diurnal rhythms of the physiological signals and a linear regression model to control for demographic factors. The cosinor analysis revealed significant differences in the mesor (rhythm-adjusted mean) of EDA, skin temperature, and heart rate between the depressed/anxious group and the control group. The linear model confirmed the association between elevated EDA and depression scores, even after controlling for demographics, physical activity, and other physiological measures.

The study concludes that consumer smartwatches can effectively capture physiological signals related to mental health in real-world settings. The findings suggest that ambulatory EDA, particularly tonic skin conductance, may be a practical and useful tool for monitoring and assessing mental health symptoms. However, the authors acknowledge limitations such as high comorbidity between depression and anxiety, the relatively mild symptom severity of the participants, and the sensor's deactivation during sleep, which affects data quality during nighttime and early morning hours.

Research Impact and Future Directions

This research makes a valuable contribution to digital mental health by demonstrating the feasibility of using consumer smartwatches to measure physiological correlates of depression and anxiety in everyday life. The finding of elevated skin conductance, particularly in the early morning, offers a potential new avenue for monitoring and managing these conditions. However, it's crucial to interpret these results cautiously due to the study's limitations, especially the high comorbidity between depression and anxiety, the relatively mild symptom severity within the sample, and the technical constraints of the wearable sensor used.

The study's observational design, while enabling real-world data collection, inherently limits the ability to draw strong causal conclusions. The observed correlation between elevated skin conductance and depression does not definitively prove a causal link. Other factors, not measured in this study, could contribute to both heightened physiological arousal and depressive symptoms. Furthermore, the sensor's deactivation during sleep introduces uncertainty about the true extent of the early morning effect, a key finding that requires further investigation with continuous monitoring technologies.

Despite these limitations, the study's findings have important implications for future research and practice. The ability to passively collect physiological data using readily available wearables opens exciting possibilities for large-scale monitoring, early detection, and personalized interventions for mental health. Future studies should focus on replicating these findings with more diverse samples, exploring the physiological differences between distinct mental health conditions, and developing algorithms that can translate real-world sensor data into actionable insights for individuals and clinicians.

Critical Analysis and Recommendations

Clear and Concise Summary of Key Information (written-content)
The abstract effectively communicates the study's objective, methodology, and key findings in a clear and organized manner. This allows readers to quickly grasp the essence of the research.
Section: ABSTRACT
Quantify Primary Finding for Increased Impact (written-content)
Quantifying the primary finding of elevated skin conductance by including effect sizes or p-values would strengthen the abstract's impact and allow readers to immediately gauge the magnitude of the effect.
Section: ABSTRACT
Clear Justification for the Study (written-content)
The introduction effectively establishes the research gap by highlighting the limitations of lab-based EDA studies and the need for real-world data.
Section: INTRODUCTION
Expand Hypotheses to Include All Measured Variables (written-content)
Including directional hypotheses for heart rate and skin temperature, in addition to EDA, would align the introduction with the study's full scope and improve the narrative flow.
Section: INTRODUCTION
Transparent Data Processing Procedures (written-content)
The methods section provides detailed information about data processing, enhancing transparency and reproducibility. This allows other researchers to scrutinize and potentially replicate the study.
Section: METHODS
Justify Data Filtering Choices for Enhanced Rigor (written-content)
Providing a rationale for the chosen signal filtering thresholds for SCL and HR would further strengthen the methodology and allow for better evaluation of the data processing steps.
Section: METHODS
Robust Validation of Control Group Data (written-content)
Comparing the control group's data to a large independent dataset validates the baseline and strengthens the credibility of the findings by demonstrating that the control group's physiological patterns align with population norms.
Section: RESULTS AND DISCUSSION
Reconcile 'Early Morning' Finding with Data Limitations (written-content)
The discussion should address the discrepancy between the 'early morning' finding and the sensor's deactivation during sleep. Clarifying the specific time window of the effect or acknowledging the data gap would improve the interpretation of this key result.
Section: RESULTS AND DISCUSSION
Improve Clarity of Figure 1 (graphical-figure)
Figure 1 effectively visualizes the diurnal patterns of physiological and behavioral measures, but the unclear y-axis label and missing panel label could be improved for better clarity.
Section: RESULTS AND DISCUSSION
Robust Population Comparison Enhances Findings (graphical-figure)
Figure 2 provides valuable context by comparing the study's findings to a large population dataset, strengthening the interpretation of the results.
Section: RESULTS AND DISCUSSION
Clear and Impactful Summary of Findings (written-content)
The conclusion effectively summarizes the key findings and their practical implications for using wearable sensors in mental health monitoring.
Section: CONCLUSION
Reinforce the Study's Novel Contribution (written-content)
Explicitly reiterating the study's core contribution of bridging the gap between lab-based research and real-world application would strengthen the conclusion's impact.
Section: CONCLUSION

Section Analysis

ABSTRACT

Key Aspects

Strengths

Suggestions for Improvement

INTRODUCTION

Key Aspects

Strengths

Suggestions for Improvement

METHODS

Key Aspects

Strengths

Suggestions for Improvement

RESULTS AND DISCUSSION

Key Aspects

Strengths

Suggestions for Improvement

Non-Text Elements

Figure 1. Diurnal patterns in physiological and behavioural measures across the...
Full Caption

Figure 1. Diurnal patterns in physiological and behavioural measures across the data set.

Figure/Table Image (Page 4)
Figure 1. Diurnal patterns in physiological and behavioural measures across the data set.
First Reference in Text
Figure 1 shows average HR, HRV RMSSD, SCL, skin temperature and step counts as a function of the depression, anxiety and stress groupings.
Description
  • Data availability varies significantly throughout the day: This line graph illustrates data availability across a 24-hour cycle. The number of participants providing sensor data is lowest during the night, dropping below 1,000 observations between approximately 01:00 and 06:00. Data availability rises sharply in the morning and remains high throughout the day, peaking at approximately 2,800 observations around 20:00.
Scientific Validity
  • ✅ Transparently displays data availability: Presenting the data availability is a commendable act of transparency. It allows the reader to critically assess the robustness of the findings presented in panel B, particularly during the early morning hours where data is sparse. This is crucial for interpreting the diurnal patterns correctly.
Communication
  • ✅ Effective use of shading to indicate data sparsity: The grey shaded area from 01:00 to 06:00 is an excellent visual device to explicitly denote the period of low data availability and thus lower confidence in the observed patterns. This directly links to the information presented in panel B, enhancing the overall coherence of the figure.
  • 💡 Y-axis label is unclear: The y-axis label, 'No. Participant Minutes,' is ambiguous. The caption clarifies that the total possible is 6636 (28 days × 237 participants), suggesting the y-axis represents the total number of participant-days with data available at each specific minute of the day. A more intuitive label, such as 'Number of Daily Observations' or 'Total Participant-Days with Data,' would improve clarity and reduce potential misinterpretation.
  • 💡 Missing panel label: The plot lacks a panel label (e.g., 'A') to clearly distinguish it from the grid of plots below it (panel B), as referenced in the caption. Adding a label would improve navigational clarity within the multi-pane figure.
Table 1 Non-linear function (CircaCompare²) outputs when comparing not...
Full Caption

Table 1 Non-linear function (CircaCompare²) outputs when comparing not depressed (PHQ-8 <5) and mildly depressed or depressed groups (PHQ- 8 ≥5), not anxious (GAD-7 <5) and mildly anxious or anxious groups (GAD-7 ≥5), and low stress (PSS <14) and moderate-to-high stress (PSS ≥14) groups.

Figure/Table Image (Page 5)
Table 1 Non-linear function (CircaCompare²) outputs when comparing not depressed (PHQ-8 <5) and mildly depressed or depressed groups (PHQ- 8 ≥5), not anxious (GAD-7 <5) and mildly anxious or anxious groups (GAD-7 ≥5), and low stress (PSS <14) and moderate-to-high stress (PSS ≥14) groups.
First Reference in Text
In our cohort of 237 subjects, we observe elevated SCL levels in the depression (vs low depression) group (see table 1 for cosinor fitting results).
Description
  • Overview of the table's purpose and methodology: This table presents a statistical comparison of daily biological rhythms between individuals with low versus high symptoms of depression (PHQ-8), anxiety (GAD-7), and stress (PSS). The analysis breaks down each 24-hour rhythm into three key components: the mesor (the rhythm's average value), the amplitude (the size of the daily fluctuation), and the phase (the timing of the daily peak).
  • Key findings for depression (PHQ-8): For individuals with higher depression symptoms (PHQ-8 ≥5), the analysis found statistically significant increases in the average daily level (mesor) of Skin Conductance Level (SCL), a measure of sweat activity, by 0.73 units; skin temperature by 0.12°C; and heart rate by 2.81 beats per minute, compared to the non-depressed group. No significant differences were found in the amplitude or timing of these rhythms.
  • Key findings for anxiety (GAD-7): Similarly, individuals with higher anxiety symptoms (GAD-7 ≥5) showed significantly higher average daily SCL (an increase of 0.70 units), skin temperature (an increase of 0.20°C), and heart rate (an increase of 2.41 beats per minute) compared to the non-anxious group.
  • Key findings for stress (PSS): The differences for perceived stress were not statistically significant at the p<0.05 level. The only notable finding was a borderline increase in the average daily skin temperature for the high-stress group (an increase of 0.09°C), which had a p-value less than 0.1.
Scientific Validity
  • ✅ Appropriate statistical method for rhythmic data: The use of cosinor analysis is highly appropriate for the research question, as it allows for a nuanced examination of diurnal patterns beyond simple daily averages. Decomposing the rhythms into mesor, amplitude, and phase provides a more detailed understanding of how physiological patterns differ between groups.
  • ✅ Inclusion of confidence intervals enhances rigor: The inclusion of 95% confidence intervals for every estimate is a methodological strength. It provides readers with a clear sense of the precision of the estimates and the potential range of the true effect sizes, which is more informative than p-values alone.
  • 💡 Potential issue with multiple comparisons: The table presents the results of 45 independent statistical tests (5 variables x 3 parameters x 3 conditions). There is no mention of an adjustment for multiple comparisons (e.g., Bonferroni correction, FDR). Without such an adjustment, the risk of finding a significant result by chance (a Type I error) is inflated. The authors should address this, either by applying a correction or by justifying its omission.
  • 💡 Lack of exact p-values: The table does not report exact p-values, instead using bolding for p<0.05 and a symbol for p<0.1. While this indicates significance, providing exact p-values would offer more granular information, distinguishing between results that are marginally significant (e.g., p=0.049) and those that are highly significant (e.g., p<0.001).
  • 💡 Potential for Type II errors due to low statistical power: The wide confidence intervals for many of the non-significant parameters (e.g., the difference in HRV RMSSD mesor for PHQ-8 is -13.79 to 7.19) suggest that the study may have been underpowered to detect smaller, yet potentially meaningful, differences. The authors should be cautious in their interpretation, framing these as 'no detected difference' rather than 'no difference'.
Communication
  • ✅ Clear and logical structure: The table is well-organized with a clear hierarchical structure, grouping results by mental health condition (PHQ-8, GAD-7, PSS), rhythmic parameter, and physiological variable. This logical layout allows readers to efficiently locate and compare specific findings.
  • ✅ Effective use of bolding for significance: Using bold text to highlight statistically significant results (p<0.05) is an effective visual cue that immediately draws the reader's attention to the most important findings of the analysis.
  • 💡 Over-reliance on abbreviations and symbols: The table relies heavily on abbreviations (SCL, HRV RMSSD) and symbols for cosinor parameters (k1, alpha1, psi1) that may not be intuitive. While a footnote defines the mental health scales, adding parenthetical explanations within the table itself (e.g., 'Diff. in mesor (average)', 'SCL (Skin Conductance)') would greatly improve its self-contained clarity and readability without requiring the reader to hunt for definitions.
  • 💡 Redundant column headers could be streamlined: The column headers 'Est.' and '95% CI' are repeated for each of the five physiological measures, creating visual redundancy. Consider using a multi-level header where each measure (e.g., 'SCL', 'Skin temp.') spans two sub-columns for 'Estimate' and '95% CI'. This would streamline the table's appearance and reduce clutter.
Table 2 Fixed effect estimates, SEs and estimated p values from the linear...
Full Caption

Table 2 Fixed effect estimates, SEs and estimated p values from the linear mixed effects model

Figure/Table Image (Page 5)
Table 2 Fixed effect estimates, SEs and estimated p values from the linear mixed effects model
First Reference in Text
Based on figure 1 differences appear greatest from 06:00 to noon; however, due to the smaller number of samples during the 06:00-07:00 period, there is also larger variance, Differences in SCL during the night-time/early morning hours between people with no depressive/anxiety symptoms and elevated depressive/ anxiety symptoms deserve further research as our results suggest there might be phenotypical differences. Elevated tonic SCL in the depressed group is observed despite their taking no more steps, suggesting that the increased sympathetic arousal may be driven by factors other than physical activity. For context, figure 2 shows a comparison between the groups in our study and baseline diurnal pattern from a sample of 15349 Fitbit Sense 2 users (for whom there is no survey data) reported in prior work. The control group (no depression, no anxiety, low stress) diurnal data in our study resemble the population level traces closely, particularly for SCL and heart rate measurement. Our results do not support theories that electrodermal hypore- activity is a feature of depression in all contexts. Differences in physiological diurnal patterns were observed when comparing PHQ score, GAD score and PSS score in our cohort of 237 subjects. As expected, and consistent with high rates of comorbidity in the literature, in this population there is a considerable overlap in subjects who were depressed and anxious in particular (see covariance matrix in online supple- mental table S2) and therefore similarity in the diurnal patterns is unsurprising. However, tonic SCL was not significantly greater for the anxious group, consistent with prior work analysing responses in lab conditions. When we consider the other physiological variables, we see patterns that confirm results in prior work. Skin temperature was slightly elevated in the depressed group, consistent with a recent large study in ambulatory settings of 20000 subjects. The result was consistent in the cosinor model and the OLS model; however, the effects were smaller than for SCL. Heart rate was higher in the depressed group in the cosinor model; however, the tonic effect was no longer significant when controlling for other variables in the OLS model. Table 2 Fixed effect estimates, SEs and estimated p values from the linear mixed effects model Predictor variable PHQ-8 GAD-7 PSS Intercept 3.41 4.51 13.9 Demographics Age -0.83 -1.15 -1.43 Weight 0.47 0.002 -0.55 Gender (female) 1.57 0.94 1.88 Orientation (LGBTQI+) 3.38 1.09 2.34 Race (not Caucasian) -0.81 -0.93 -0.44 Phys./behaviour Mean tonic SCL 0.23 0.03 0.24 Mean HR 0.06 0.09 0.39 Mean HRV RMSSD -0.15 -0.28 -0.63 Mean skin temp. 0.05 0.12 0.06 Mean steps -0.09 0.13 0.02 Bold values significant to p<0.01. p<0.01. GAD, Generalised Anxiety Disorder; HR, Heart Rate; HRV, Heart Rate Variability; LGBTQI+, Lesbian, gay, bisexual, transgender, queer, intersex, asexual, and other related identities.; PHQ, Patient Health Questionnaire; PSS, Perceived Stress Scale; RMSSD, root mean square of successive differences; SCL, skin conductance level.
Description
  • Overview of the statistical model and its outputs: This table presents the results of a linear mixed effects model, a statistical analysis designed to understand the relationship between various factors and mental health scores in a group of individuals over time. The numbers in the table, called 'fixed effect estimates', show how much a mental health score (for depression, anxiety, or stress) is predicted to change for every one-unit increase in a given predictor, after accounting for all other factors in the model.
  • Key physiological predictors of depression scores: For depression (PHQ-8), higher scores are significantly associated with higher mean tonic Skin Conductance Level (SCL), which measures sweat activity as an indicator of arousal (coefficient = 0.23), and higher mean heart rate (0.06). Conversely, higher Heart Rate Variability (HRV RMSSD), a measure of beat-to-beat changes in heart rate often linked to better stress regulation, is associated with lower depression scores (-0.15).
  • Key physiological and behavioral predictors of anxiety scores: For anxiety (GAD-7), several factors show a significant association. Notably, higher mean heart rate (0.09), higher skin temperature (0.12), and more steps taken (0.13) are linked to higher anxiety scores. In contrast, higher HRV RMSSD is associated with lower anxiety scores (-0.28).
  • Key physiological and demographic predictors of stress scores: For stress (PSS), the model shows that higher mean tonic SCL (0.24) is linked to higher stress scores, while higher HRV RMSSD is associated with lower stress scores (-0.63). Among demographics, being female (1.88) and identifying as LGBTQI+ (2.34) were associated with higher stress scores, while older age (-1.43) and higher weight (-0.55) were associated with lower scores.
Scientific Validity
  • ✅ Appropriate statistical model for longitudinal data: The use of a linear mixed effects model is a significant strength. This approach is well-suited for longitudinal data as it correctly accounts for the non-independence of repeated measurements within the same individual, providing more accurate and reliable estimates than a standard regression model.
  • ✅ Comprehensive model controls for multiple confounders: The model simultaneously includes demographic, physiological, and behavioral predictors. This multivariate approach allows for the examination of the unique contribution of each variable while controlling for others, which is a more rigorous method than the univariate comparisons presented in Table 1.
  • 💡 Inconsistency in the description of the statistical model: The reference text mentions using an 'OLS model', while the methods section and this table's caption refer to a 'linear mixed effects model'. These are different statistical techniques. This inconsistency should be resolved to ensure clarity and accuracy regarding the methodology used.
  • 💡 Potential for inflated Type I error due to multiple testing: The table reports coefficients for multiple predictors for three different outcomes. Given the number of statistical tests performed, there is an increased risk of Type I errors (false positives). The authors should clarify whether any correction for multiple comparisons was applied or provide a justification for its absence.
Communication
  • ✅ Logical grouping of predictor variables: The table's structure is clear, with predictor variables logically grouped into 'Demographics' and 'Phys./behaviour'. This organization makes it easy for the reader to navigate the results and compare the effects of different types of variables.
  • 💡 Caption is inconsistent with table content: The caption is inconsistent with the table's content. It states that the table includes 'SEs and estimated p values', but neither standard errors nor exact p-values are presented. The table only shows the coefficient estimates and uses an asterisk to indicate significance. The caption should be revised to accurately describe the contents.
  • 💡 Conflicting significance indicators in the footnote: The footnote is confusing as it provides two different indicators for the same significance level: 'Bold values significant to p<0.01' and '*p<0.01'. Since only asterisks are used in the table body, the reference to bold values should be removed to avoid ambiguity and improve clarity.
  • 💡 Missing units for predictor variables: The units of the predictor variables are not provided. For example, it is unclear if physiological measures are in their raw units or are standardized (Z-scored, as mentioned in the methods). This omission makes it difficult to interpret the practical magnitude of the coefficients. Adding a note about units or standardization would significantly improve the table's interpretability.
Figure 2 Comparison between study diurnal patterns (n=237) and population...
Full Caption

Figure 2 Comparison between study diurnal patterns (n=237) and population diurnal patterns (n=15349).

Figure/Table Image (Page 6)
Figure 2 Comparison between study diurnal patterns (n=237) and population diurnal patterns (n=15349).
First Reference in Text
Based on figure 1 differences appear greatest from 06:00 to noon; however, due to the smaller number of samples during the 06:00-07:00 period, there is also larger variance, Differences in SCL during the night-time/early morning hours between people with no depressive/anxiety symptoms and elevated depressive/ anxiety symptoms deserve further research as our results suggest there might be phenotypical differences. Elevated tonic SCL in the depressed group is observed despite their taking no more steps, suggesting that the increased sympathetic arousal may be driven by factors other than physical activity. For context, figure 2 shows a comparison between the groups in our study and baseline diurnal pattern from a sample of 15349 Fitbit Sense 2 users (for whom there is no survey data) reported in prior work.
Description
  • Overview of the multi-panel comparison: This figure displays five line graphs, each showing the 24-hour pattern of a different physiological or behavioral measure: Heart Rate, Heart Rate Variability (HRV RMSSD), Skin Conductance Level (SCL), Skin Temperature, and Steps per Minute. It compares three groups: a large general population sample (n=15,349), and two subgroups from the main study (n=237)—those without depression and those with depression.
  • Elevated physiological arousal in the depression group: The group with depression consistently shows higher heart rate, higher skin temperature, and higher Skin Conductance Level (SCL) throughout the day compared to the non-depressed group and the general population. SCL, a measure of sweat gland activity that reflects sympathetic nervous system arousal, shows a particularly pronounced difference in the morning hours, where the depressed group's SCL is around 4-5 µSiemens, while the other groups are closer to 2-3 µSiemens.
  • Reduced Heart Rate Variability in the depression group: The depressed group exhibits lower Heart Rate Variability (HRV RMSSD) across the 24-hour period. HRV RMSSD measures the tiny variations in time between consecutive heartbeats and is often considered an indicator of the body's ability to adapt to stress. The depressed group's HRV is consistently about 10 ms lower than the non-depressed group's.
  • Similar physical activity levels between study groups: Despite the clear differences in physiological measures, the daily pattern of physical activity, measured as steps per minute, is nearly identical for the depressed and non-depressed groups within the study. This suggests that the observed physiological differences are not simply due to one group being more physically active than the other.
  • Study's control group is representative of the general population: The physiological patterns of the 'No Depression' study group closely mirror those of the much larger 'Gen. Population' sample across all metrics. This indicates that the non-depressed participants in the study are representative of the broader population, strengthening their validity as a control group.
Scientific Validity
  • ✅ Use of a large population sample for context: Including a very large (n=15,349) general population sample as a baseline is a significant methodological strength. It provides a robust, real-world context for the study's findings and increases confidence that the 'No Depression' group is a suitable control.
  • ✅ Effectively controls for physical activity as a confounder: By presenting physiological data alongside behavioral data (steps), the figure strongly supports the paper's argument that the elevated sympathetic arousal in the depressed group is not merely a byproduct of different physical activity levels. This is a crucial control that strengthens the study's conclusions.
  • 💡 Lack of uncertainty intervals: The figure presents only the mean trend lines without any indication of variance or uncertainty (e.g., shaded 95% confidence intervals or standard error bands). Without this information, it is impossible to visually assess the statistical significance of the differences between the groups. The observed gaps could fall within the natural variability of the data.
  • 💡 Unknown composition of the general population sample: The reference text states the 'Gen. Population' group has no associated survey data. This is a limitation, as its mental health composition is unknown. It likely includes individuals with varying levels of depression and anxiety, which would mean the true physiological profile of a 'healthy' population might differ from what is shown. This makes the comparison between the 'Depression' group and the general population less direct than it appears.
Communication
  • ✅ Consistent and clear color coding: The use of a consistent and distinct color scheme for the three groups across all five panels is highly effective. It allows for quick and intuitive comparison of the different physiological measures for the same group.
  • ✅ Effective multi-panel grid layout: The grid layout, where each physiological measure is plotted in a separate panel against a common time axis, is an excellent choice for presenting this complex dataset. It facilitates easy comparison of the diurnal patterns across different metrics.
  • 💡 Legend could be more informative: The legend is clear but could be enhanced by including the sample sizes for each group (e.g., 'Gen. Population (n=15,349)', 'No Depression (n=...)', 'Depression (n=...)'). This would make the figure more self-contained and immediately highlight the vast difference in scale between the population and study samples.
  • 💡 Acronyms are not defined in the caption: While the caption is informative, it omits the full name for 'RMSSD'. To improve the figure's self-sufficiency, spell out key acronyms either in the caption or as a footnote, for instance, 'Root Mean Square of Successive Differences (RMSSD)'.

CONCLUSION

Key Aspects

Strengths

Suggestions for Improvement